]> bbs.cooldavid.org Git - net-next-2.6.git/blame - kernel/rcutree.h
uids: Prevent tear down race
[net-next-2.6.git] / kernel / rcutree.h
CommitLineData
9f77da9f
PM
1/*
2 * Read-Copy Update mechanism for mutual exclusion (tree-based version)
3 * Internal non-public definitions.
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18 *
19 * Copyright IBM Corporation, 2008
20 *
21 * Author: Ingo Molnar <mingo@elte.hu>
22 * Paul E. McKenney <paulmck@linux.vnet.ibm.com>
23 */
24
25#include <linux/cache.h>
26#include <linux/spinlock.h>
27#include <linux/threads.h>
28#include <linux/cpumask.h>
29#include <linux/seqlock.h>
30
31/*
32 * Define shape of hierarchy based on NR_CPUS and CONFIG_RCU_FANOUT.
33 * In theory, it should be possible to add more levels straightforwardly.
34 * In practice, this has not been tested, so there is probably some
35 * bug somewhere.
36 */
37#define MAX_RCU_LVLS 3
38#define RCU_FANOUT (CONFIG_RCU_FANOUT)
39#define RCU_FANOUT_SQ (RCU_FANOUT * RCU_FANOUT)
40#define RCU_FANOUT_CUBE (RCU_FANOUT_SQ * RCU_FANOUT)
41
42#if NR_CPUS <= RCU_FANOUT
43# define NUM_RCU_LVLS 1
44# define NUM_RCU_LVL_0 1
45# define NUM_RCU_LVL_1 (NR_CPUS)
46# define NUM_RCU_LVL_2 0
47# define NUM_RCU_LVL_3 0
48#elif NR_CPUS <= RCU_FANOUT_SQ
49# define NUM_RCU_LVLS 2
50# define NUM_RCU_LVL_0 1
fc2219d4 51# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT)
9f77da9f
PM
52# define NUM_RCU_LVL_2 (NR_CPUS)
53# define NUM_RCU_LVL_3 0
54#elif NR_CPUS <= RCU_FANOUT_CUBE
55# define NUM_RCU_LVLS 3
56# define NUM_RCU_LVL_0 1
fc2219d4
PM
57# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_SQ)
58# define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT)
9f77da9f
PM
59# define NUM_RCU_LVL_3 NR_CPUS
60#else
61# error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
62#endif /* #if (NR_CPUS) <= RCU_FANOUT */
63
64#define RCU_SUM (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3)
65#define NUM_RCU_NODES (RCU_SUM - NR_CPUS)
66
67/*
68 * Dynticks per-CPU state.
69 */
70struct rcu_dynticks {
71 int dynticks_nesting; /* Track nesting level, sort of. */
72 int dynticks; /* Even value for dynticks-idle, else odd. */
73 int dynticks_nmi; /* Even value for either dynticks-idle or */
74 /* not in nmi handler, else odd. So this */
75 /* remains even for nmi from irq handler. */
76};
77
78/*
79 * Definition for node within the RCU grace-period-detection hierarchy.
80 */
81struct rcu_node {
1eba8f84
PM
82 spinlock_t lock; /* Root rcu_node's lock protects some */
83 /* rcu_state fields as well as following. */
f41d911f 84 long gpnum; /* Current grace period for this node. */
86848966
PM
85 /* This will either be equal to or one */
86 /* behind the root rcu_node's gpnum. */
9f77da9f
PM
87 unsigned long qsmask; /* CPUs or groups that need to switch in */
88 /* order for current grace period to proceed.*/
1eba8f84
PM
89 /* In leaf rcu_node, each bit corresponds to */
90 /* an rcu_data structure, otherwise, each */
91 /* bit corresponds to a child rcu_node */
92 /* structure. */
9f77da9f
PM
93 unsigned long qsmaskinit;
94 /* Per-GP initialization for qsmask. */
95 unsigned long grpmask; /* Mask to apply to parent qsmask. */
1eba8f84 96 /* Only one bit will be set in this mask. */
9f77da9f
PM
97 int grplo; /* lowest-numbered CPU or group here. */
98 int grphi; /* highest-numbered CPU or group here. */
99 u8 grpnum; /* CPU/group number for next level up. */
100 u8 level; /* root is at level 0. */
101 struct rcu_node *parent;
f41d911f
PM
102 struct list_head blocked_tasks[2];
103 /* Tasks blocked in RCU read-side critsect. */
1eba8f84
PM
104 /* Grace period number (->gpnum) x blocked */
105 /* by tasks on the (x & 0x1) element of the */
106 /* blocked_tasks[] array. */
9f77da9f
PM
107} ____cacheline_internodealigned_in_smp;
108
a0b6c9a7
PM
109/*
110 * Do a full breadth-first scan of the rcu_node structures for the
111 * specified rcu_state structure.
112 */
113#define rcu_for_each_node_breadth_first(rsp, rnp) \
114 for ((rnp) = &(rsp)->node[0]; \
115 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
116
117#define rcu_for_each_leaf_node(rsp, rnp) \
118 for ((rnp) = (rsp)->level[NUM_RCU_LVLS - 1]; \
119 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
120
9f77da9f
PM
121/* Index values for nxttail array in struct rcu_data. */
122#define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
123#define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
124#define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
125#define RCU_NEXT_TAIL 3
126#define RCU_NEXT_SIZE 4
127
128/* Per-CPU data for read-copy update. */
129struct rcu_data {
130 /* 1) quiescent-state and grace-period handling : */
131 long completed; /* Track rsp->completed gp number */
132 /* in order to detect GP end. */
133 long gpnum; /* Highest gp number that this CPU */
134 /* is aware of having started. */
135 long passed_quiesc_completed;
136 /* Value of completed at time of qs. */
137 bool passed_quiesc; /* User-mode/idle loop etc. */
138 bool qs_pending; /* Core waits for quiesc state. */
139 bool beenonline; /* CPU online at least once. */
f41d911f 140 bool preemptable; /* Preemptable RCU? */
9f77da9f
PM
141 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
142 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
143
144 /* 2) batch handling */
145 /*
146 * If nxtlist is not NULL, it is partitioned as follows.
147 * Any of the partitions might be empty, in which case the
148 * pointer to that partition will be equal to the pointer for
149 * the following partition. When the list is empty, all of
1eba8f84
PM
150 * the nxttail elements point to the ->nxtlist pointer itself,
151 * which in that case is NULL.
9f77da9f 152 *
9f77da9f
PM
153 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
154 * Entries that batch # <= ->completed
155 * The grace period for these entries has completed, and
156 * the other grace-period-completed entries may be moved
157 * here temporarily in rcu_process_callbacks().
1eba8f84
PM
158 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
159 * Entries that batch # <= ->completed - 1: waiting for current GP
160 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
161 * Entries known to have arrived before current GP ended
162 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
163 * Entries that might have arrived after current GP ended
164 * Note that the value of *nxttail[RCU_NEXT_TAIL] will
165 * always be NULL, as this is the end of the list.
9f77da9f
PM
166 */
167 struct rcu_head *nxtlist;
168 struct rcu_head **nxttail[RCU_NEXT_SIZE];
a71fca58 169 long qlen; /* # of queued callbacks */
37c72e56
PM
170 long qlen_last_fqs_check;
171 /* qlen at last check for QS forcing */
172 unsigned long n_force_qs_snap;
173 /* did other CPU force QS recently? */
9f77da9f
PM
174 long blimit; /* Upper limit on a processed batch */
175
176#ifdef CONFIG_NO_HZ
177 /* 3) dynticks interface. */
178 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
179 int dynticks_snap; /* Per-GP tracking for dynticks. */
180 int dynticks_nmi_snap; /* Per-GP tracking for dynticks_nmi. */
181#endif /* #ifdef CONFIG_NO_HZ */
182
183 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
184#ifdef CONFIG_NO_HZ
185 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
186#endif /* #ifdef CONFIG_NO_HZ */
187 unsigned long offline_fqs; /* Kicked due to being offline. */
188 unsigned long resched_ipi; /* Sent a resched IPI. */
189
190 /* 5) __rcu_pending() statistics. */
191 long n_rcu_pending; /* rcu_pending() calls since boot. */
192 long n_rp_qs_pending;
193 long n_rp_cb_ready;
194 long n_rp_cpu_needs_gp;
195 long n_rp_gp_completed;
196 long n_rp_gp_started;
197 long n_rp_need_fqs;
198 long n_rp_need_nothing;
199
200 int cpu;
201};
202
203/* Values for signaled field in struct rcu_state. */
204#define RCU_GP_INIT 0 /* Grace period being initialized. */
205#define RCU_SAVE_DYNTICK 1 /* Need to scan dyntick state. */
206#define RCU_FORCE_QS 2 /* Need to force quiescent state. */
207#ifdef CONFIG_NO_HZ
208#define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK
209#else /* #ifdef CONFIG_NO_HZ */
210#define RCU_SIGNAL_INIT RCU_FORCE_QS
211#endif /* #else #ifdef CONFIG_NO_HZ */
212
213#define RCU_JIFFIES_TILL_FORCE_QS 3 /* for rsp->jiffies_force_qs */
214#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
215#define RCU_SECONDS_TILL_STALL_CHECK (10 * HZ) /* for rsp->jiffies_stall */
216#define RCU_SECONDS_TILL_STALL_RECHECK (30 * HZ) /* for rsp->jiffies_stall */
217#define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */
218 /* to take at least one */
219 /* scheduling clock irq */
220 /* before ratting on them. */
221
222#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
223
224/*
225 * RCU global state, including node hierarchy. This hierarchy is
226 * represented in "heap" form in a dense array. The root (first level)
227 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
228 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
229 * and the third level in ->node[m+1] and following (->node[m+1] referenced
230 * by ->level[2]). The number of levels is determined by the number of
231 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
232 * consisting of a single rcu_node.
233 */
234struct rcu_state {
235 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
236 struct rcu_node *level[NUM_RCU_LVLS]; /* Hierarchy levels. */
237 u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */
238 u8 levelspread[NUM_RCU_LVLS]; /* kids/node in each level. */
239 struct rcu_data *rda[NR_CPUS]; /* array of rdp pointers. */
240
241 /* The following fields are guarded by the root rcu_node's lock. */
242
243 u8 signaled ____cacheline_internodealigned_in_smp;
244 /* Force QS state. */
245 long gpnum; /* Current gp number. */
246 long completed; /* # of last completed gp. */
1eba8f84
PM
247
248 /* End of fields guarded by root rcu_node's lock. */
249
9f77da9f 250 spinlock_t onofflock; /* exclude on/offline and */
e74f4c45
PM
251 /* starting new GP. Also */
252 /* protects the following */
253 /* orphan_cbs fields. */
254 struct rcu_head *orphan_cbs_list; /* list of rcu_head structs */
255 /* orphaned by all CPUs in */
256 /* a given leaf rcu_node */
257 /* going offline. */
258 struct rcu_head **orphan_cbs_tail; /* And tail pointer. */
259 long orphan_qlen; /* Number of orphaned cbs. */
9f77da9f
PM
260 spinlock_t fqslock; /* Only one task forcing */
261 /* quiescent states. */
262 unsigned long jiffies_force_qs; /* Time at which to invoke */
263 /* force_quiescent_state(). */
264 unsigned long n_force_qs; /* Number of calls to */
265 /* force_quiescent_state(). */
266 unsigned long n_force_qs_lh; /* ~Number of calls leaving */
267 /* due to lock unavailable. */
268 unsigned long n_force_qs_ngp; /* Number of calls leaving */
269 /* due to no GP active. */
270#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
271 unsigned long gp_start; /* Time at which GP started, */
272 /* but in jiffies. */
273 unsigned long jiffies_stall; /* Time at which to check */
274 /* for CPU stalls. */
275#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
276#ifdef CONFIG_NO_HZ
277 long dynticks_completed; /* Value of completed @ snap. */
278#endif /* #ifdef CONFIG_NO_HZ */
279};
280
281#ifdef RCU_TREE_NONCORE
6258c4fb
IM
282
283/*
284 * RCU implementation internal declarations:
285 */
d6714c22
PM
286extern struct rcu_state rcu_sched_state;
287DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
6258c4fb
IM
288
289extern struct rcu_state rcu_bh_state;
290DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
291
f41d911f
PM
292#ifdef CONFIG_TREE_PREEMPT_RCU
293extern struct rcu_state rcu_preempt_state;
294DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
295#endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
296
9b2619af 297#else /* #ifdef RCU_TREE_NONCORE */
9f77da9f 298
9b2619af
PM
299/* Forward declarations for rcutree_plugin.h */
300static inline void rcu_bootup_announce(void);
301long rcu_batches_completed(void);
302static void rcu_preempt_note_context_switch(int cpu);
303static int rcu_preempted_readers(struct rcu_node *rnp);
304#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
305static void rcu_print_task_stall(struct rcu_node *rnp);
306#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
307static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
308#ifdef CONFIG_HOTPLUG_CPU
237c80c5
PM
309static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
310 struct rcu_node *rnp,
311 struct rcu_data *rdp);
9b2619af
PM
312static void rcu_preempt_offline_cpu(int cpu);
313#endif /* #ifdef CONFIG_HOTPLUG_CPU */
314static void rcu_preempt_check_callbacks(int cpu);
315static void rcu_preempt_process_callbacks(void);
316void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
317static int rcu_preempt_pending(int cpu);
318static int rcu_preempt_needs_cpu(int cpu);
319static void __cpuinit rcu_preempt_init_percpu_data(int cpu);
e74f4c45 320static void rcu_preempt_send_cbs_to_orphanage(void);
9b2619af
PM
321static void __init __rcu_init_preempt(void);
322
323#endif /* #else #ifdef RCU_TREE_NONCORE */